[e2e] on local ethernet throughput?

[David P. Reed]

At 12:42 PM 10/29/2001 -0800, Christian Huitema wrote:
>David, I think you are wrong on this one. The carriers were under an FCC
>requirement to allow customers to choose their own ISP. That pretty much
>mandates some form of switching "below IP" so that the bits sent by the
>DSL modem are delivered to the chosen ISP, regardless of the value of
>the destination IP address.

I must not be being clear.  Of course there needs to be a backhaul from the 
access DSL line to an ISP's point of presence.  And it certainly can be 
"below IP".  But ATM circuits (especially *subscriber-initiated* ATM 
circuits, rather than nailed-up circuits) are a bad choice.  And I think 
this choice was driven primarily by vendors' definition of what a DSLAM 
is.  In particular ATM creates the notion that there is a "cost" associated 
with a "connected" ATM circuit that does not have to be borne by when the 
ATM circuit is idle.  Leading to the fiction that one can charge for 
"minutes" of ATM connectivity, and discourage holding circuits open 24x7, 
which would create true "always on" IP connectivity.

>As Craig said, ATM makes some sense; there
>are alternatives such as frame relay and MPLS, or PPPoE for that matter;
>whether the alternatives are better or worse is debatable. You can only
>use "straight IP" if you accept that whoever provides the media
>connectivity also provides the IP service, which may or may not be a
>desirable outcome.

The framing protocol for backhaul can be anything you want.   The 
connectivity provider need not (and as you say, probably should not) 
provide the IP service.  They just provide a link.  The key is not to 
confuse the carriers into thinking that these are "phone calls" - they are 
permanent connections for the duration of the subscriber's contract.


- David
--------------------------------------------
WWW Page: http://www.reed.com/dpr.html